Abstract
Earth Observation satellite imaging helps building diagnosis during a disaster. Several models are put forward on the xBD dataset, which can be divided into two levels: the building level and the pixel level. Models from two levels evolve into several versions that will be reviewed in this paper. There are four key challenges hindering researchers from moving forward on this task, and this paper tries to give technical solutions. First, metrics on different levels could not be compared directly. We put forward a fairer metric and give a method to convert between metrics of two levels. Secondly, drone images may be another important source, but drone data may have only a post-disaster image. This paper shows and compares methods of directly detecting and generating. Thirdly, the class imbalance is a typical feature of the xBD dataset and leads to a bad F1 score for minor damage and major damage. This paper provides four specific data resampling strategies, which are Main-Label Over-Sampling (MLOS), Discrimination After Cropping (DAC), Dilation of Area with Minority (DAM) and Synthetic Minority Over-Sampling Technique (SMOTE), as well as cost-sensitive re-weighting schemes. Fourthly, faster prediction meets the need for a real-time situation. This paper recommends three specific methods, feature-map subtraction, parameter sharing, and knowledge distillation. Finally, we developed our AI-driven Damage Diagnose Platform (ADDP). This paper introduces the structure of ADDP and technical details. Customized settings, interface preview, and upload and download satellite images are major services our platform provides.
Highlights
Motivation and Problem StatementNatural disasters such as floods, hurricanes, or earthquakes cause great loss of life, property damage, and economic damage every year around the world
For building-level tasks [7,9,10,11,12,13,14], each building has a ground-truth polygon that outlines its location in the image
We introduce 3 specific methods: Main-Label Over-Sampling (MLOS), Discrimination After Cropping (DAC) and Dilation of Area with Minority (DAM)
Summary
Natural disasters such as floods, hurricanes, or earthquakes cause great loss of life, property damage, and economic damage every year around the world. A series of research results has been achieved regarding convolutional neural network-based building-damage assessment from satellite imagery, there are still many challenges to be discussed. We combine our experience and experiments with the use of the xBD dataset to contribute to the following: First, we made a comprehensive state-of-the-art review of convolutional neural network-based Building-Damage Assessment from Satellite Imagery. We conducted a technical discussion for the four key challenges of operational convolutional neural network-based building-damage assessment from satellite imagery as detailed below:. (2) Challenge 2: How Do We Conduct Building-Damage Assessment in the Absence of Pre-Disaster Satellite Imagery?. For pre-disaster imagery, a three-band RGB image and building polygons are provided. For post-disaster imagery, a three-band RGB image and building classifications based on The Joint Damage Scale are provided. Different disaster types or sensor types differ in the imagery of the building damage
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
